An Estimation Of Distance Of An Imaged Target In Video Sequence

The imaging sequence of a target is the foundation of video based tracking, which has been finding its wide application in target monitoring, air defense, traffic management, and in many other fields. Especially, distance estimation of a non-cooperative target is a technical difficulty in imaging detection but has important application value. With technical requirements of the National Natural Science Foundation of China funded project "Research on target’s distance estimation system and the key technologies based on monocular image and direction (No.60872136)" this paper mainly research on the distance estimation algorithm optimization, real-time implementation technologies as well as its automatic robust implementation.This paper firstly introduces the target’s distance estimation model and related algorithm based on feature segment. Main points include getting more than three SIFT key matching points between two adjacent images, selecting three matching points to construct a virtual circle for the preparation of the target’s feature segment in it, and solving a special ranging equation to get distance. High-performance DSP was used to meet the real-time requirement in our work, but the algorithm’s complexity adds the difficulty in its real-time implement.Secondly, to reduce time consuming of the algorithm we simplified the SIFT algorithm as far as possible. On the other hand, combined with the characteristics of C64x+DSP core, a lot of optimization strategies such as using intermediate variables, intrinsic operations, word access to short type data, the compiler settings, software pipelining and linear assembly have been used to optimize target’s distance estimation algorithm program. Also, related optimization based on DSP’s storage structure has been made. Finally, the target’s distance estimation algorithm achieved the real-time requirement of25frames per second.Lastly, the practical study of our method has been done under the condition of low contrast and low visual image. Image contrast enhancement algorithm such as the piecewise linearization method and logarithmic transformation to image gray level was used to improve the adaptability of our whole algorithm. An image sharpening method using logarithmic to enhance target’s image contrast was also used. The experimental result shows that this method can make target’s distance estimation algorithm well adapts to target’s maneuverability. Considering that an image may suffer many kinds of noises and disturbs in its generation and transmission, for minimizing these affection, research on robust application technology for target distance estimation algorithm is also carried out. Applications of all these methods have enhanced target’s distance estimation algorithm’s practicability as well as its robustness.